Machine learning-based surveying and mapping geographic space full-coverage data generation method and system

Abstract

The invention relates to the technical field of geographic information science, in particular to a surveying and mapping geographic space full-coverage data generation method and system based on machine learning. The method comprises the following steps: firstly, collecting a historical forest vegetation image, a season label, climate data and a real vegetation coverage image required by a vegetation coverage image generation model; secondly, a historical forest vegetation image analysis model and a climate data analysis model are constructed, historical forest vegetation images and climate data are analyzed respectively, and forest vegetation spatial features and climate time features are obtained; thirdly, fusing and processing the forest vegetation spatial features, the climate time features and the season labels to obtain and generate a vegetation coverage image; then, constructing a generated vegetation coverage image authenticity scoring model, and calculating the difference between the authenticity and the generated vegetation coverage image to obtain a generated vegetation coverage image authenticity score; and finally, optimizing the authenticity of the generated vegetation coverage image by using the adversarial loss, and obtaining the generated vegetation coverage image meeting a preset condition.

Description

Technical Field

[0001] The present invention relates to the field of geographic information science and technology, and specifically to a method and system for generating full-coverage data of surveying and mapping geospatial space based on machine learning. Background Art

[0002] Full geospatial coverage data refers to a collection of geographic data covering an entire geographic area. Such datasets, typically including topographic maps, satellite imagery, and aerial photographs, provide comprehensive geographic information, providing fundamental data support for various geographic analyses and applications. These full-coverage geographic datasets are widely used in fields such as urban planning, resource management, and environmental monitoring. However, collecting full geospatial coverage data in complex terrain remains challenging, particularly in areas with large terrain gradients and complex spatial patterns of vegetation. Optical remote sensing imagery is often limited by cloud cover, making it difficult to obtain continuous, cloud-free imagery within the same season. Furthermore, in natural transition zones, the inherent ambiguity of vegetation classification and interference from human activities make the detection of seasonal changes uncertain. These difficulties and uncertainties significantly complicate the monitoring and simulation of seasonal changes in forest vegetation, hindering research progress in ecological research, resource management, and disaster warning. Therefore, a data generation method is urgently needed to generate forest vegetation data with spatiotemporal characteristics.

[0003] Machine learning has been successfully applied to tasks such as image generation and image restoration. However, forest vegetation growth not only depends on diverse terrains, soils, and water sources, but also on long-term climate patterns and sudden climate events. Therefore, machine learning can struggle to capture spatial and temporal dependencies when generating forest vegetation cover data.

[0004] To this end, a method and system for generating full coverage data of surveying and mapping geospatial space based on machine learning is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for generating full-coverage data for surveying and mapping geographic space based on machine learning. First, historical forest vegetation images, seasonal labels, climate data, and real vegetation coverage images required for a vegetation coverage image generation model are collected; second, a historical forest vegetation image analysis model and a climate data analysis model are constructed to analyze the historical forest vegetation images and climate data, respectively, to obtain forest vegetation spatial characteristics and a climate event matrix; then, the forest vegetation spatial characteristics, climate time characteristics, and seasonal label vectors are fused and processed to obtain a generated vegetation coverage image; then, a generated vegetation coverage image authenticity scoring model is constructed to calculate the difference between the real and generated vegetation coverage images to obtain a generated vegetation coverage image authenticity score; finally, an adversarial loss is used to continuously optimize the authenticity of the generated vegetation coverage image to obtain a generated vegetation coverage image that meets preset conditions.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The method for generating full coverage data of surveying and mapping geospatial space based on machine learning includes:

[0008] Collect historical remote sensing image data of the forest at different time points to obtain historical forest vegetation data; the historical forest vegetation data includes: historical forest vegetation images and season labels; collect time series data related to climate to obtain climate data; collect actual forest vegetation cover images to obtain real vegetation cover images;

[0009] According to the historical forest vegetation image, the climate data and the season label, a vegetation cover image generation model is constructed to generate a vegetation cover image; the specific steps are:

[0010] Constructing a historical forest vegetation image analysis model, analyzing the historical forest vegetation image, and obtaining forest vegetation spatial characteristics;

[0011] Constructing a climate data analysis model, analyzing the climate data, and obtaining climate time characteristics;

[0012] fusing the forest vegetation spatial features, the climate temporal features, and the season label vector to obtain conditional features;

[0013] Processing the conditional features to obtain the generated vegetation cover image;

[0014] Constructing a authenticity scoring model for the generated vegetation cover image, and obtaining an authenticity score for the generated vegetation cover image by calculating the difference between the real vegetation cover image and the generated vegetation cover image in combination with the conditional features;

[0015] According to the authenticity score of the generated vegetation coverage image, the authenticity of the generated vegetation coverage image is continuously optimized using adversarial loss to obtain the generated vegetation coverage image that meets preset conditions.

[0016] Preferably, the historical forest vegetation data includes: historical forest vegetation images and season labels; wherein the historical forest vegetation images include: historical NDVI images; the season labels include: spring, summer, autumn and winter;

[0017] The climate data includes temperature, precipitation, humidity and sunshine duration.

[0018] Preferably, the vegetation cover image generation model is constructed through a CGAN model, including: a generator, a discriminator and conditional data; the generator includes: the historical forest vegetation image analysis model and the climate data analysis model; the discriminator includes: the generated vegetation cover image authenticity scoring model; the conditional data includes: the historical forest vegetation image, the climate data and the season label.

[0019] Preferably, the historical forest vegetation image analysis model is constructed by a CNN model, comprising: a first input layer, a spatial feature extraction layer and a first output layer; the analysis process of the historical forest vegetation image is:

[0020] The first input layer inputs the historical forest vegetation image into the historical forest vegetation image analysis model;

[0021] The spatial feature extraction layer uses the CNN model to extract the spatial features of the historical forest vegetation image;

[0022] The first output layer outputs the spatial features extracted by the spatial feature extraction layer to obtain the forest vegetation spatial features.

[0023] Preferably, the climate data analysis model is constructed by an LSTM model, including: a second input layer, a time feature extraction layer, and a second output layer; the analysis process of the climate data is:

[0024] The second input layer inputs the climate data into the climate data analysis model;

[0025] The time feature extraction layer extracts the time features of the climate data through a double-layer LSTM model;

[0026] The second output layer outputs the time features extracted by the time feature extraction layer to obtain the climate time features.

[0027] Preferably, the forest vegetation spatial features, the climate time features and the season label vector are nonlinearly processed and fused to obtain the conditional features; the conditional features are processed to obtain the generated vegetation coverage image; the specific process includes: a feature fusion layer and a vegetation coverage image generation layer;

[0028] The feature fusion layer fuses the forest vegetation spatial features, the climate temporal features, and the season label vector through a fully connected layer to obtain the conditional features. The formula for the conditional features is:

[0029] F fusion =σ1(ω1·g1(NDVI)+ω2·g2(f climate )+ω3·g3(fseason )+ω4·g4(NDVI,f climate ,f season ))+b1;

[0030] Among them, F fusion is the conditional characteristic; NDVI is the spatial characteristic of forest vegetation; f climate is the climate time characteristic; f season is the seasonal label vector; ω1 is the weight of the spatial characteristics of forest vegetation; g1 is the nonlinear transformation function of the spatial characteristics of forest vegetation; ω2 is the weight of the climate event characteristics; g2 is the nonlinear transformation function of the climate time characteristics; ω3 is the weight of the seasonal label vector; g3 is the nonlinear transformation function of the seasonal label vector; ω4 is the weight of the interaction term of the spatial characteristics of forest vegetation, the climate time characteristics and the seasonal label vector; g4 is the nonlinear transformation function of the interaction term of the spatial characteristics of forest vegetation, the climate time characteristics and the seasonal label vector; σ1 is the activation function; b1 is the bias term;

[0031] Combining the conditional features with noise to obtain vegetation cover image features;

[0032] The vegetation coverage image generation layer uses deconvolution to generate a high-resolution image according to the vegetation coverage image features to obtain the generated vegetation coverage image.

[0033] Preferably, the authenticity scoring model for generating vegetation coverage images includes: calculating the forest vegetation spatial feature differences, climate time feature differences, and seasonal label differences between the real vegetation coverage image and the generated vegetation coverage image, as well as the interaction term differences between the three, to obtain the authenticity score of the generated vegetation coverage image; the formula for the authenticity score of the generated vegetation coverage image is:

[0034]

[0035] S=σ2·S1+b2;

[0036] Among them, S1 is the difference between the real vegetation cover image and the generated vegetation cover image; NDVI real The spatial characteristics of forest vegetation are real vegetation cover images; NDVI gen To generate forest vegetation spatial characteristics for vegetation cover images; f climate,real is the climate characteristics of the real vegetation cover image; f climate,gen To generate climate characteristics of vegetation cover images; f season,real is the season label of the real vegetation cover image; f season,genis the seasonal label for generating vegetation cover images; α1 is the weight of the difference in forest vegetation spatial characteristics; h1 is the difference in forest vegetation spatial characteristics; α2 is the weight of the difference in climate time characteristics; h2 is the difference in climate time characteristics; α3 is the weight of the difference in seasonal labels; h3 is the difference in seasonal labels; α4 is the weight of the interaction term among forest vegetation spatial characteristics, climate time characteristics and seasonal label vectors; h4 is the difference in the interaction term among forest vegetation spatial characteristics, climate time characteristics and seasonal label vectors; S is the authenticity score; σ2 is the activation function; and b2 is the bias term.

[0037] Preferably, the preset condition is that the authenticity score of the generated vegetation coverage image is higher than an authenticity score threshold and the generated vegetation coverage image meets conditional data.

[0038] A machine learning-based system for generating full-coverage geospatial surveying and mapping data, including:

[0039] The data acquisition module is used to collect historical remote sensing image data of the forest at different time points to obtain historical forest vegetation data; the historical forest vegetation data includes: historical forest vegetation images and season labels; collect time series data related to climate to obtain climate data; collect actual forest vegetation cover images to obtain real vegetation cover images;

[0040] The vegetation cover image generation module is used to construct a vegetation cover image generation model based on the historical forest vegetation image, the climate data and the season label to generate a vegetation cover image; the specific steps are:

[0041] Constructing a historical forest vegetation image analysis model, analyzing the historical forest vegetation image, and obtaining forest vegetation spatial characteristics;

[0042] Constructing a climate data analysis model, analyzing the climate data, and obtaining climate time characteristics;

[0043] Nonlinearly processing and fusing the forest vegetation spatial characteristics, the climate temporal characteristics, and the season label vector to obtain conditional characteristics;

[0044] Processing the conditional features to obtain the generated vegetation cover image;

[0045] A generated vegetation cover image authenticity scoring module is used to construct a generated vegetation cover image authenticity scoring model, and obtain a generated vegetation cover image authenticity score by calculating the difference between the generated vegetation cover image and the real vegetation cover image in combination with the conditional features;

[0046] The generated vegetation cover image optimization module is used to continuously optimize the authenticity of the generated vegetation cover image using adversarial loss according to the authenticity score of the generated vegetation cover image, so as to obtain the generated vegetation cover image that meets preset conditions.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. The present invention proposes a historical forest vegetation image analysis model, which uses a CNN model to extract spatial features of historical forest vegetation images, captures local textures, shapes, and global spatial distribution features in the image, and generates a feature matrix that reflects the distribution patterns of vegetation types and changes in vegetation density. The feature matrix can serve as the input and constraint of the generator, provide a historical reference for the generated vegetation coverage image, help the generator learn the vegetation distribution patterns, and ensure that the generated vegetation coverage image is consistent with historical data in spatial distribution. At the same time, this method can retain the continuity and consistency of the geographic space, avoid unreasonable distribution phenomena in the generated image, such as isolated patches and discontinuous strip distributions, thereby improving the authenticity and ecological significance of the generated vegetation coverage image; combining the spatial features extracted by the model with the temporal features extracted by the subsequent climate data analysis model can improve the authenticity of the vegetation coverage image to be generated by the present invention.

[0049] 2. This paper proposes a climate data analysis model that uses an LSTM model to extract temporal features from climate data. This model can capture the temporal dynamics of climate data, including short-term fluctuations, seasonal patterns, and long-term trends. A two-layer LSTM network processes the input climate data step by step, extracting latent features at each time step and capturing the temporal dependencies of climate factors through a memory and forgetting mechanism. The resulting temporal feature matrix output by the LSTM not only retains key climate data information but also integrates contextual information across time steps. These temporal feature matrices are used as input to a generator to simulate the dynamic response of vegetation to climate change, reflecting seasonal patterns such as vegetation recovery due to spring warming and lush vegetation growth due to high temperatures and high humidity in summer. This method of extracting temporal features from climate data through LSTM not only provides accurate temporal dynamic information for vegetation cover image generation but also enhances the logical consistency, ecological rationality, and responsiveness of the generated results to diverse climate conditions. This method is a key step in generating seasonal vegetation cover images and helps ensure the image generation capabilities of subsequent vegetation cover image generation models.

[0050] 3. The present invention proposes a authenticity scoring model for generating vegetation cover images. By comparing the generated vegetation cover images with the real images and calculating the differences, the authenticity scoring model can quantify the consistency between the generated images and the real data in terms of spatial distribution, texture characteristics, and seasonal variation patterns. At the same time, the authenticity scoring model combines conditional features to calculate the differences between the generated vegetation cover images and the real vegetation cover images, ensuring that the generated vegetation cover images are not only visually close to the real vegetation cover images, but also conform to climatic, geographical, and ecological laws. The optimization of the adversarial loss guides the generator to learn features that are closer to the real images, making the generated results more realistic and reducing the occurrence of false features or unreasonable structures, thereby improving the authenticity of the generated vegetation cover images. Through the optimization of the authenticity scoring model and the adversarial loss, the generated vegetation cover images achieve higher quality in terms of vision, ecology, and application. This method not only improves the authenticity of image generation, but also ensures its scientific rationality and diversity under a variety of complex conditions, providing strong support for the research and practice of ecological models. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A flowchart of a method for generating full coverage data of surveying and mapping geospatial space based on machine learning provided in an embodiment of the present invention;

[0052] Figure 2 A structural diagram of a system for generating full-coverage surveying and mapping geographic space data based on machine learning provided by an embodiment of the present invention;

[0053] Figure 3 A flowchart of generating a vegetation cover image provided by an embodiment of the present invention;

[0054] Figure 4 The embodiment of the present invention provides a method for generating a vegetation cover image authenticity score map. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0056] Full geospatial coverage data refers to a collection of geographic data covering an entire geographic area. Such datasets, typically including topographic maps, satellite imagery, and aerial photographs, provide comprehensive geographic information, providing fundamental data support for various geographic analyses and applications. These full-coverage geographic datasets are widely used in fields such as urban planning, resource management, and environmental monitoring. However, collecting full geospatial coverage data in complex terrain remains challenging, particularly in areas with large terrain gradients and complex spatial patterns of vegetation. Optical remote sensing imagery is often limited by cloud cover, making it difficult to obtain continuous, cloud-free imagery within the same season. Furthermore, in natural transition zones, the inherent ambiguity of vegetation classification and interference from human activities make the detection of seasonal changes uncertain. These difficulties and uncertainties significantly complicate the monitoring and simulation of seasonal changes in forest vegetation, hindering research progress in ecological research, resource management, and disaster warning. Therefore, a data generation method is urgently needed to generate forest vegetation data with spatiotemporal characteristics.

[0057] The present invention proposes a method for generating full coverage data of surveying and mapping geospatial space based on machine learning, which realizes the generation of seasonal forest vegetation cover data. This method is applied to the system for generating full coverage data of surveying and mapping geospatial space based on machine learning. For the specific method and system flowchart, please refer to Figure 1 and Figure 2 In order to illustrate that the method and system of the present invention can generate vegetation cover data, the effectiveness of the present invention will be described below using two embodiments.

[0058] See also Figures 1 to 4 The present invention provides a method and system for generating full coverage data of surveying and mapping geospatial space based on machine learning. The technical solution is as follows:

[0059] Example 1

[0060] In the embodiment of the present application, the method and system proposed by the present invention are used to describe in detail the process of generating a forest vegetation cover image. In the embodiment of the present application, the generation of the forest vegetation cover image is aimed at the vegetation cover image of a forest area A for one year. Figure 1 and Figure 2 The content describes in detail the process of generating the vegetation cover image of the forest area A; Figure 1The specific process of the method proposed in the present invention includes: first, collecting historical forest vegetation images, season labels, climate data and real vegetation cover images required for a vegetation cover image generation model; second, constructing a historical forest vegetation image analysis model to analyze the historical forest vegetation images to obtain forest vegetation spatial characteristics; then, constructing a climate data analysis model to analyze the climate data to obtain climate time characteristics; third, fusing and processing the forest vegetation spatial characteristics, climate time characteristics and season label vectors to obtain a generated vegetation cover image; then, constructing a generated vegetation cover image authenticity scoring model to calculate the difference between the real vegetation cover image and the generated vegetation cover image to obtain a generated vegetation cover image authenticity score; finally, using adversarial loss to continuously optimize the authenticity of the generated vegetation cover image to obtain a generated vegetation cover image that meets preset conditions. Figure 2 The structural diagram of the system proposed in the present invention includes: a data acquisition module, a vegetation cover image generation module, a vegetation cover image authenticity scoring module and a vegetation cover image optimization module; Figure 1 and Figure 2 The following describes the contents:

[0061] Collect historical remote sensing image data of Forest A at different time points through MODIS to obtain historical forest vegetation data; the historical forest vegetation data includes historical forest vegetation images and corresponding season labels; the historical forest vegetation images include historical NDVI images; the season labels include spring, summer, autumn and winter;

[0062] The historical forest vegetation images were spatially and temporally aligned to ensure uniform resolution; the effects of noise such as cloud cover and shadows were removed through filtering; and the NDVI value range was standardized;

[0063] The climate data of Forest A is obtained by collecting climate-related time series data from weather station records. The climate data includes temperature, precipitation, humidity, and sunshine duration. The temperature includes monthly average temperature and extreme temperature. The precipitation includes monthly cumulative precipitation. The humidity includes average relative humidity. The sunshine duration includes monthly cumulative sunshine duration.

[0064] Filling missing records in the climate data using interpolation; aligning the climate data with the time steps of the historical forest vegetation data;

[0065] The actual vegetation cover image of Forest A is collected through high-resolution remote sensing images to obtain a real vegetation cover image;

[0066] Cropping the real vegetation cover image to a size consistent with the resolution of the historical NDVI image;

[0067] In an embodiment of the present application, by collecting historical forest vegetation images and corresponding season labels, it is possible to record the forest vegetation coverage in different seasons, reflect the spatial distribution, density and health status of the vegetation, and capture the law of vegetation changes with the seasons; the historical NDVI image serves as the core input of the subsequent historical forest vegetation image analysis model, providing the spatial characteristics of the vegetation and providing data support for the generation of vegetation coverage images. By collecting climate data, the relationship between vegetation changes and climate conditions can be revealed; it provides conditional input for the subsequent climate data analysis model to capture the dynamic relationship of vegetation changes over time, which can ensure that the generated vegetation coverage image reflects the vegetation distribution under a specific climate scenario. By collecting real vegetation coverage images, it can be used to compare the differences between the generated images and the real data, guide the generator to learn vegetation features that are closer to the real distribution, and ensure that the generated vegetation coverage image is more realistic.

[0068] Preferably, a vegetation cover image generation model is constructed based on the historical forest vegetation image, the climate data and the season label to generate a vegetation cover image. Figure 3 Flowchart of vegetation cover image generation;

[0069] The vegetation cover image generation model is constructed using a CGAN model, and includes: a generator, a discriminator, and conditional data; the generator includes: the historical forest vegetation image analysis model and the climate data analysis model; the discriminator includes: the authenticity scoring model for the generated vegetation cover image; the conditional data includes: the historical forest vegetation image, the climate data, and the season label:

[0070] c={c ndvi ,c climate ,c season};

[0071] Among them, c is the conditional data; c ndvi is the historical forest vegetation image; c climate is climate data; c season For season labels.

[0072] In an embodiment of the present application, a generator in a vegetation cover image generation model generates realistic images of seasonal changes in forest vegetation based on input historical forest vegetation images and climate data, combined with seasonal labels. A discriminator in the vegetation cover image generation model determines whether the vegetation cover image generated by the generator is realistic. By embedding conditional data into the generator and the discriminator and fusing the input data, the model's ability to discriminate data consistency is improved. By constructing a vegetation cover image generation model, the goal of generating realistic vegetation cover images can be achieved.

[0073] Preferably, a historical forest vegetation image analysis model is constructed to analyze the historical forest vegetation image to obtain forest vegetation spatial features; the historical forest vegetation image analysis model is constructed using a CNN model, and the historical forest vegetation image analysis model includes: a first input layer, a spatial feature extraction layer, and a first output layer; the analysis process of the historical forest vegetation image is:

[0074] The first input layer inputs the historical forest vegetation image into the historical forest vegetation image analysis model;

[0075] The spatial feature extraction layer uses a CNN model to extract the spatial features of the historical forest vegetation image; the CNN model is a two-layer CNN network structure: the first layer includes 64 3×3 convolution kernels, followed by a 2×2 maximum pooling layer; the second layer includes 128 3×3 convolution kernels, followed by global average pooling to reduce the dimensionality of the spatial features to a one-dimensional vector;

[0076] The first output layer outputs the spatial features extracted by the spatial feature extraction layer to obtain the forest vegetation spatial features:

[0077] NVDI=CNN(c ndvi );

[0078] Among them, NVDI is the spatial characteristic of forest vegetation.

[0079] The embodiment of the present application proposes a historical forest vegetation image analysis model to extract spatial features from historical forest vegetation images. The spatial feature extraction through the CNN model can effectively identify and learn the local and global spatial features in the historical forest vegetation images. For example, CNN can extract information such as the shape, distribution density, texture, etc. of the vegetation area from historical remote sensing images, thereby generating a vegetation coverage image with natural transitions and realism. By extracting spatial features from the image, the generator of the cGAN model can capture the spatial distribution patterns of vegetation in different seasons and time points, and further optimize the generation process. These spatial features will help the generator create high-quality images that conform to actual forest vegetation changes, so that the generated vegetation coverage images have higher authenticity and accuracy.

[0080] Preferably, a climate data analysis model is constructed to analyze the climate data to obtain climate time characteristics; the climate data analysis model is constructed by an LSTM model, and the climate data analysis model includes: a second input layer, a time feature extraction layer, and a second output layer; the analysis process of the climate data is:

[0081] The second input layer inputs the climate data into the climate data analysis model;

[0082] The temporal feature extraction layer extracts the temporal features of the climate data through a double-layer LSTM, where each LSTM layer contains 128 units and uses Dropout to prevent overfitting;

[0083] The second output layer outputs the time feature extracted by the time feature extraction layer to obtain the climate time feature:

[0084] f climate =LSTM(c climate );

[0085] Among them, f climate Temporal characteristics of climate.

[0086] The embodiment of the present application proposes a climate data analysis model to extract temporal features from climate data. In the process of generating vegetation cover images, the LSTM model can help the generator understand the evolution of climate change over time. For example, the LSTM model can learn the time series patterns of climate factors such as temperature, precipitation, and humidity, and extract the impact of these factors on vegetation growth. Through the training of the two-layer LSTM model, the temporal information of climate data is effectively captured and converted into climate time features, providing dynamically changing climate conditions for the generation model. These temporal features can guide the generator to simulate the growth and seasonal changes of vegetation under different climatic conditions, and generate more accurate and time-dependent vegetation cover images, thereby improving the authenticity and reliability of the generated vegetation cover images.

[0087] Preferably, the forest vegetation spatial features, the climate time features and the season label vector are fused to obtain conditional features; the conditional features are processed to obtain the generated vegetation coverage image; the specific process includes: a feature fusion layer and a vegetation coverage image generation layer;

[0088] The feature fusion layer performs nonlinear processing on the forest vegetation spatial features, the climate temporal features, and the season label vector and fuses them to obtain the conditional features; the season label vector is mapped into a continuous vector through the embedding layer:

[0089] f season =Embedding(c season );

[0090] Among them, f season is the season label vector;

[0091] The formula of the conditional feature is:

[0092] F fusion =σ1(ω1·g1(NDVI)+ω2·g2(f climate )+ω3·g3(fseason )+ω4·g4(NDVI,f climate ,f season ))+b1;

[0093] g1(NDVI)=β1·NDVI+β2·NDVI 2 ;

[0094]

[0095] g4(NVDI,f climate ,f season )=f1·NVDI·T avg +f2·NVDI·P total +f3·NVDI·f season ;

[0096] Among them, F fusion is the conditional characteristic; NDVI is the spatial characteristic of forest vegetation; f climate is the climate time characteristic; f season is the seasonal label vector; ω1 is the weight of the nonlinear processing of forest vegetation spatial characteristics; g1 is the nonlinear transformation function of forest vegetation spatial characteristics; ω2 is the weight of the nonlinear processing of climate event characteristics; g2 is the nonlinear transformation function of climate time characteristics; ω3 is the weight of the nonlinear processing of season label vector; g3 is the nonlinear transformation function of season label vector; ω4 is the weight of the nonlinear processing of the interaction term of forest vegetation spatial characteristics, climate time characteristics and seasonal label vector; g4 is the nonlinear transformation function of the interaction term of forest vegetation spatial characteristics, climate time characteristics and seasonal label vector; σ1 is the activation function; b1 is the bias term; β1 is the weight of forest vegetation spatial characteristics; β2 is the weight of the secondary effect of forest vegetation spatial characteristics; c1 is the weight of monthly average temperature; T avg is the monthly mean temperature; c2 is the weight of nonlinear processing of monthly total precipitation; P total is the monthly total precipitation; c3 is the weight of the nonlinear processing of the monthly average humidity; H avg is the monthly average humidity; c4 is the weight of the nonlinear processing of the monthly total sunshine duration; S total is the total monthly sunshine duration; e1 is the weight of the season label vector; f1 is the weight of the interaction term between the spatial characteristics of forest vegetation and the monthly average temperature; f2 is the weight of the interaction term between the spatial characteristics of forest vegetation and the total monthly precipitation; f3 is the weight of the interaction term between the spatial characteristics of forest vegetation and the season label;

[0097] The conditional features are combined with noise to obtain vegetation cover image features; the specific formula is:

[0098] f fusion =FC(z,f cond );

[0099] Among them, f fusion is the vegetation cover image feature; z is random noise;

[0100] The vegetation coverage image generation layer uses deconvolution to generate a high-resolution image according to the vegetation coverage image features to obtain the generated vegetation coverage image; the generation formula of the vegetation coverage image is:

[0101] NDVI gen =G(f fusion );

[0102] in, To generate vegetation coverage images; G is the generator network.

[0103] In an embodiment of the present application, the spatial characteristics of forest vegetation, the climate-time characteristics and the season label vector are nonlinearly processed and fused, so that the generated vegetation coverage image can be more consistent with the actual seasonal changes. The spatial characteristics of forest vegetation reflect the vegetation types and distribution in different regions, the climate-time characteristics reflect the changes in climatic factors such as temperature and precipitation, and the season label indicates the current season. By fusing this information, the generated vegetation coverage image can accurately simulate the changes in forests under different seasons and climatic conditions, and show the real growth process. For example, in spring, the generated image may show more greenery, while in winter it may appear as sparse vegetation or covered with snow. In this way, the vegetation coverage image finally generated not only has a reasonable spatial layout, but also conforms to the natural laws of time and climate, making it more realistic.

[0104] Preferably, a authenticity scoring model for the generated vegetation cover image is constructed, and the authenticity score of the generated vegetation cover image is obtained by calculating the difference between the generated vegetation cover image and the real vegetation cover image in combination with the conditional features; the authenticity scoring model for the generated vegetation cover image includes: calculating the difference in forest vegetation spatial characteristics, climate time characteristics and seasonal labels between the real vegetation cover image and the generated vegetation cover image, as well as the difference in the interaction term between the three, to obtain the authenticity score of the generated vegetation cover image; the formula for the authenticity score of the generated vegetation cover image is:

[0105]

[0106] S=σ2·S1+b2;

[0107] h1(NDVI real ,NDVI gen )=k1·(NDVI real -NDVI gen ) 2 +k2·|NDVIreal -NDVI gen |;

[0108]

[0109] h4(NDVI,f climate ,f season )=n1·NDVI·T avg +n2·NDVI·P total +n3·NDVI·S avg +n4·P total ·S avg ;

[0110] Among them, S1 is the difference between the real vegetation cover image and the generated vegetation cover image; NDVI real The spatial characteristics of forest vegetation are real vegetation cover images; NDVI gen To generate forest vegetation spatial characteristics for vegetation cover images; f climate,real is the climate characteristics of the real vegetation cover image; f climate,gen To generate climate characteristics of vegetation cover images; f season,real is the season label of the real vegetation cover image; f season,gen is the seasonal label for generating vegetation cover images; α1 is the weight of the difference in forest vegetation spatial characteristics; h1 is the difference in forest vegetation spatial characteristics; α2 is the weight of the difference in climate time characteristics; h2 is the difference in climate time characteristics; α3 is the weight of the difference in seasonal labels; h3 is the difference in seasonal labels; α4 is the weight of the interaction term of forest vegetation spatial characteristics, climate time characteristics and seasonal label vectors; h4 is the difference in the interaction term of forest vegetation spatial characteristics, climate time characteristics and seasonal label vectors; S is the authenticity score; σ2 is the activation function; b2 is the bias term; k1 is the weight of the quadratic effect of forest vegetation spatial characteristics difference; k2 is the weight of the forest vegetation spatial characteristics difference value; T real is the actual monthly mean temperature; T gen is the monthly average temperature; l1 is the temperature difference weight; P real is the actual monthly total precipitation; P gen is the monthly total precipitation; l2 is the precipitation difference weight; H real is the actual monthly average humidity; H gen is the monthly average humidity; l3 is the humidity difference weight; S real is the actual total monthly sunshine duration; S genis used to generate the total monthly sunshine duration; l4 is the sunshine duration difference weight; m is the seasonal label difference weight; n1 is the weight of the interaction term between the spatial characteristics of forest vegetation and the monthly average temperature; n2 is the weight of the interaction term between the spatial characteristics of forest vegetation and the monthly total precipitation; n3 is the weight of the interaction term between the spatial characteristics of forest vegetation and the monthly average humidity; n4 is the weight of the interaction term between the monthly total precipitation and the monthly average humidity.

[0111] The embodiment of the present application proposes a model for scoring the authenticity of generated vegetation coverage images. By calculating the difference between the generated vegetation coverage image and the real vegetation coverage image, the quality and authenticity of the generated image can be effectively evaluated. This step can ensure that the generated vegetation coverage image can be as close as possible to the vegetation coverage status of the real forest. For example, the generated vegetation coverage image may vary due to factors such as climate and seasonal changes. The authenticity scoring model gives an evaluation value by comparing the differences between the generated image and the actually observed vegetation image (such as color, texture, and distribution, etc.). If the score is low, it means that the generated image may not be realistic enough and the generator needs to be further optimized; conversely, a high score indicates that the generated vegetation image is already very close to the actual situation. This evaluation process helps to improve the quality of image generation so that it can accurately reflect the actual situation of the forest in different application scenarios, such as environmental monitoring and agricultural forecasting.

[0112] Preferably, the authenticity of the generated vegetation coverage image is continuously optimized by using adversarial loss to obtain the generated vegetation coverage image that meets preset conditions;

[0113] Specifically, the process of continuously optimizing the authenticity of the generated vegetation cover image using adversarial loss includes: discriminator update, generator update and iterative optimization;

[0114] The discriminator updating includes: fixing the generator, and updating the parameters of the discriminator so that the discriminator can accurately classify the real vegetation cover image and the generated vegetation cover image;

[0115] The generator update includes: fixing the discriminator and updating the generator parameters so that the generated vegetation cover image gradually meets preset conditions and approaches the real vegetation cover image; the preset conditions are: the authenticity score of the generated vegetation cover image is higher than the authenticity score threshold, and the generated vegetation cover image meets the condition data;

[0116] The iterative optimization includes: through alternating training of the generator and the discriminator, by minimizing the generator loss and the generator, the authenticity of the generated vegetation coverage image is continuously improved, and finally a realistic generated vegetation coverage image is obtained.

[0117] See also Figure 4, by training a large number of real and generated vegetation images and optimizing the weight value, we can get Figure 4 The authenticity score graph of the generated vegetation coverage image is shown in Figure 2; Table 1 shows the authenticity score and adversarial loss of the generated vegetation coverage image during the iterative optimization process.

[0118] Table 1. Reality score and adversarial loss of generated vegetation cover images during iterative optimization.

[0119]

[0120] As can be seen from Table 1, as the number of iterations increases, the discriminator's score for the real vegetation cover image gradually approaches 1, indicating that the discriminator has gradually improved its ability to judge the real image; the discriminator's score for the generated vegetation cover image gradually approaches 1, indicating that the discriminator gradually optimizes the authenticity of the generated vegetation cover image through alternating training of the generator and the discriminator; at the same time, the generator's loss gradually decreases, indicating that the quality of the vegetation cover image generated by the generator is improving and can better deceive the discriminator.

[0121] In an embodiment of the present application, the authenticity of the generated vegetation coverage image is continuously optimized using adversarial loss, with the main purpose of making the generated vegetation coverage image closer and closer to the real forest scene. The generator and the discriminator "compete" with each other through adversarial training: the generator continuously improves, trying to generate more and more realistic vegetation coverage images, while the discriminator continuously judges whether the image is real. In this way, the generated images will gradually approach the real forest vegetation coverage and meet preset conditions such as seasonal changes and climate influences. Ultimately, the generated images not only look more natural, but also can accurately reflect the changes in forest vegetation coverage under different environmental and time conditions, ensuring that they can better represent the actual situation in applications and be used for subsequent geospatial research.

[0122] The embodiment of the present application realizes the generation of seasonal vegetation cover images of Forest A by constructing a vegetation cover image generation model, a historical forest vegetation image analysis model, a climate data analysis model, and a vegetation cover image authenticity scoring model. First, by collecting historical remote sensing image data of the forest at different time points, the density and distribution of forest vegetation in different seasons can be intuitively reflected, laying the foundation for the subsequent extraction of spatial features; by collecting climate-related time series data, climate change information such as temperature, precipitation, humidity, and sunshine duration is obtained to provide support for analyzing the dynamic impact of climate on vegetation growth; by collecting actual vegetation cover images of the forest, the model is helped to gradually improve the generation quality during the optimization process. Secondly, by constructing a historical forest vegetation image analysis model and a climate data analysis model, a generator of a vegetation cover image generation model is constructed; the historical forest vegetation image analysis model uses a CNN model to extract the spatial features of historical images, generates forest vegetation spatial features, and ensures that the spatial distribution of the generated image conforms to the actual vegetation morphology; the climate data analysis model uses LSTM to extract the temporal features of climate data, generates climate temporal features, and captures the influence of climate changes over time on vegetation. Next, by fusing forest vegetation spatial characteristics, climate temporal characteristics, and seasonal label vectors, the spatial distribution, temporal dynamics, and seasonal characteristics are integrated to form conditional features, providing comprehensive information input for generating images. Then, by constructing a authenticity scoring model for generating vegetation cover images, the generated images are compared with the real images, and the authenticity of the generated results is quantified. The model of the present invention can adjust the generation strategy based on the scoring results. Finally, the authenticity of the generated vegetation cover images is optimized using adversarial loss. Through adversarial training of the generator and discriminator, the quality of the generated images is gradually improved. During the continuous optimization process, the generator is able to simulate more complex vegetation changes, while the discriminator ensures the authenticity of the generated vegetation cover images.

[0123] Example 2

[0124] In Example 1, the method and system proposed in the present invention successfully achieved the generation of seasonal vegetation cover images for Forest A. To further verify the effectiveness of the present invention, seasonal vegetation cover images were also generated for another forest area B in the present embodiment.

[0125] In order to better generate subsequent vegetation coverage images, an embodiment of the present invention provides a system for generating full coverage data of surveying and mapping geospatial space based on machine learning, including:

[0126] The data acquisition module is used to collect historical remote sensing image data of the forest at different time points to obtain historical forest vegetation data; the historical forest vegetation data includes: historical forest vegetation images and season labels; collect time series data related to climate to obtain climate data; collect actual forest vegetation cover images to obtain real vegetation cover images;

[0127] The historical forest vegetation data includes: historical forest vegetation images and season labels; wherein the historical forest vegetation images include: historical NDVI images; the season labels include: spring, summer, autumn and winter;

[0128] The climate data includes temperature, precipitation, humidity and sunshine duration.

[0129] Preferably, the vegetation cover image generation module is used to construct a vegetation cover image generation model based on the historical forest vegetation image, the climate data and the season label to obtain a generated vegetation cover image; the vegetation cover image generation model is constructed through a CGAN model, including: a generator, a discriminator and conditional data; the generator includes: the historical forest vegetation image analysis model and the climate data analysis model; the discriminator includes: the authenticity scoring model of the generated vegetation cover image; the conditional data includes: the historical forest vegetation image, the climate data and the season label;

[0130] A historical forest vegetation image analysis model is constructed to analyze the historical forest vegetation image to obtain forest vegetation spatial features. The historical forest vegetation image analysis model is constructed using a CNN model, comprising: a first input layer, a spatial feature extraction layer, and a first output layer. The analysis process of the historical forest vegetation image is as follows:

[0131] The first input layer inputs the historical forest vegetation image into the historical forest vegetation image analysis model;

[0132] The spatial feature extraction layer uses a CNN model to extract spatial features of the historical forest vegetation image;

[0133] The first output layer outputs the spatial features extracted by the spatial feature extraction layer to obtain the forest vegetation spatial features;

[0134] A climate data analysis model is constructed to analyze the climate data to obtain climate time characteristics. The climate data analysis model is constructed using an LSTM model and includes: a second input layer, a time feature extraction layer, and a second output layer. The analysis process of the climate data is as follows:

[0135] The second input layer inputs the climate data into the climate data analysis model;

[0136] The time feature extraction layer extracts the time features of the climate data through a double-layer LSTM;

[0137] The second output layer outputs the time feature extracted by the time feature extraction layer to obtain the climate time feature:

[0138] f climate =LSTM(c climate );

[0139] Among them, f climate Temporal characteristics of climate.

[0140] The forest vegetation spatial features, the climate time features and the season label vector are nonlinearly processed and fused to obtain conditional features; the conditional features are processed to obtain the generated vegetation cover image; the specific process includes: a feature fusion layer and a vegetation cover image generation layer;

[0141] The feature fusion layer performs nonlinear processing on the forest vegetation spatial features, the climate temporal features, and the season label vector and fuses them to obtain the conditional features; the season label vector is mapped into a continuous vector through the embedding layer:

[0142] f season =Embedding(c season );

[0143] Among them, f season is the season label vector;

[0144] The formula of the conditional feature is:

[0145] F fusion =σ1(ω1·g1(NDVI)+ω2·g2(f climate )+ω3·g3(f season )+ω4·g4(NDVI,f climate ,f season ))+b1;

[0146] g1(NDVI)=β1·NDVI+β2·NDVI 2 ;

[0147]

[0148]

[0149] g4(NVDI,f climate ,f season )=f1·NVDI·T avg +f2·NVDI·P total +f3·NVDI·f season ;

[0150] Among them, F fusion is the conditional characteristic; NDVI is the spatial characteristic of forest vegetation; f climate is the climate time characteristic; f seasonis the seasonal label vector; ω1 is the weight of the nonlinear processing of forest vegetation spatial characteristics; g1 is the nonlinear transformation function of forest vegetation spatial characteristics; ω2 is the weight of the nonlinear processing of climate event characteristics; g2 is the nonlinear transformation function of climate time characteristics; ω3 is the weight of the nonlinear processing of season label vector; g3 is the nonlinear transformation function of season label vector; ω4 is the weight of the nonlinear processing of the interaction term of forest vegetation spatial characteristics, climate time characteristics and seasonal label vector; g4 is the nonlinear transformation function of the interaction term of forest vegetation spatial characteristics, climate time characteristics and seasonal label vector; σ1 is the activation function; b1 is the bias term; β1 is the weight of forest vegetation spatial characteristics; β2 is the weight of the secondary effect of forest vegetation spatial characteristics; c1 is the weight of monthly average temperature; T avg is the monthly mean temperature; c2 is the weight of nonlinear processing of monthly total precipitation; P total is the monthly total precipitation; c3 is the weight of the nonlinear processing of the monthly average humidity; H avg is the monthly average humidity; c4 is the weight of the nonlinear processing of the monthly total sunshine duration; S total is the total monthly sunshine duration; e1 is the weight of the season label vector; f1 is the weight of the interaction term between the spatial characteristics of forest vegetation and the monthly average temperature; f2 is the weight of the interaction term between the spatial characteristics of forest vegetation and the total monthly precipitation; f3 is the weight of the interaction term between the spatial characteristics of forest vegetation and the season label;

[0151] The conditional features are combined with noise to obtain vegetation cover image features; the specific formula is:

[0152] f fusion =FC(z,f cond );

[0153] Among them, f fusion is the vegetation cover image feature; z is random noise;

[0154] The vegetation coverage image generation layer uses deconvolution to generate a high-resolution image according to the vegetation coverage image features to obtain the generated vegetation coverage image; the generation formula of the vegetation coverage image is:

[0155] NDVI gen =G(f fusion );

[0156] in, To generate vegetation coverage images; G is the generator network.

[0157] Preferably, a authenticity scoring model for the generated vegetation cover image is constructed, and the authenticity score of the generated vegetation cover image is obtained by calculating the difference between the generated vegetation cover image and the real vegetation cover image in combination with the conditional features; the authenticity scoring model for the generated vegetation cover image includes: calculating the difference in forest vegetation spatial characteristics, climate time characteristics and seasonal labels between the real vegetation cover image and the generated vegetation cover image, as well as the difference in the interaction term between the three, to obtain the authenticity score of the generated vegetation cover image; the formula for the authenticity score of the generated vegetation cover image is:

[0158]

[0159] S=σ2·S1+b2;

[0160] h1(NDVI real ,NDVI gen )=k1·(NDVI real -NDVI gen ) 2 +k2·|NDVI real -NDVI gen |;

[0161]

[0162] h4(NDVI,f climate ,f season )=n1·NDVI·T avg +n2·NDVI·P total +n3·NDVI·S avg +n4·P total ·S avg ;

[0163] Among them, S1 is the difference between the real vegetation cover image and the generated vegetation cover image; NDVI real The spatial characteristics of forest vegetation are real vegetation cover images; NDVI gen To generate forest vegetation spatial characteristics for vegetation cover images; f climate,real is the climate characteristics of the real vegetation cover image; f climate,gen To generate climate characteristics of vegetation cover images; f season,real is the season label of the real vegetation cover image; f season,genis the seasonal label for generating vegetation cover images; α1 is the weight of the difference in forest vegetation spatial characteristics; h1 is the difference in forest vegetation spatial characteristics; α2 is the weight of the difference in climate time characteristics; h2 is the difference in climate time characteristics; α3 is the weight of the difference in seasonal labels; h3 is the difference in seasonal labels; α4 is the weight of the interaction term of forest vegetation spatial characteristics, climate time characteristics and seasonal label vectors; h4 is the difference in the interaction term of forest vegetation spatial characteristics, climate time characteristics and seasonal label vectors; S is the authenticity score; σ2 is the activation function; b2 is the bias term; k1 is the weight of the quadratic effect of forest vegetation spatial characteristics difference; k2 is the weight of the forest vegetation spatial characteristics difference value; T real is the actual monthly mean temperature; T gen is the monthly average temperature; l1 is the temperature difference weight; P real is the actual monthly total precipitation; P gen is the monthly total precipitation; l2 is the precipitation difference weight; H real is the actual monthly average humidity; H gen is the monthly average humidity; l3 is the humidity difference weight; S real is the actual total monthly sunshine duration; S gen is used to generate the total monthly sunshine duration; l4 is the sunshine duration difference weight; m is the seasonal label difference weight; n1 is the weight of the interaction term between the spatial characteristics of forest vegetation and the monthly average temperature; n2 is the weight of the interaction term between the spatial characteristics of forest vegetation and the monthly total precipitation; n3 is the weight of the interaction term between the spatial characteristics of forest vegetation and the monthly average humidity; n4 is the weight of the interaction term between the monthly total precipitation and the monthly average humidity.

[0164] Preferably, a generated vegetation cover image optimization module is used to continuously optimize the authenticity of the generated vegetation cover image using adversarial loss to obtain the generated vegetation cover image that meets preset conditions; the adversarial loss includes: generator loss and discriminator loss;

[0165] The process of continuously optimizing the authenticity of the generated vegetation cover image using adversarial loss includes: discriminator update, generator update and iterative optimization;

[0166] The discriminator updating includes: fixing the generator, and updating the parameters of the discriminator so that the discriminator can accurately classify the real vegetation cover image and the generated vegetation cover image;

[0167] The generator update includes: fixing the discriminator and updating the generator parameters so that the generated vegetation cover image gradually meets preset conditions and approaches the real vegetation cover image; the preset conditions are: the authenticity score of the generated vegetation cover image is higher than the authenticity score threshold, and the generated vegetation cover image meets the condition data;

[0168] The iterative optimization includes: through alternating training of the generator and the discriminator, by minimizing the generator loss and the generator, the authenticity of the generated vegetation coverage image is continuously improved, and finally a realistic generated vegetation coverage image is obtained.

[0169] Please refer to Table 2, which shows the authenticity score and adversarial loss of the generated vegetation cover image of forest area B during the iterative optimization process.

[0170] Table 2 Authenticity scores and adversarial losses of generated vegetation cover images during iterative optimization

[0171]

[0172] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating full coverage data of surveying and mapping geospatial space based on machine learning, characterized in that: include: Collect historical remote sensing image data of forests at different time points to obtain historical forest vegetation data; The historical forest vegetation data includes: historical forest vegetation images and season labels; collecting time series data related to climate to obtain climate data; collecting actual forest vegetation cover images to obtain real vegetation cover images; According to the historical forest vegetation image, the climate data and the season label, a vegetation cover image generation model is constructed to generate a vegetation cover image; the specific steps are: Constructing a historical forest vegetation image analysis model, analyzing the historical forest vegetation image, and obtaining forest vegetation spatial characteristics; Constructing a climate data analysis model, analyzing the climate data, and obtaining climate time characteristics; Nonlinearly processing and fusing the forest vegetation spatial characteristics, the climate temporal characteristics, and the season label vector to obtain conditional characteristics; Processing the conditional features to obtain the generated vegetation cover image; Constructing a authenticity scoring model for the generated vegetation cover image, and obtaining an authenticity score for the generated vegetation cover image by calculating the difference between the real vegetation cover image and the generated vegetation cover image in combination with the conditional features; According to the authenticity score of the generated vegetation coverage image, the authenticity of the generated vegetation coverage image is continuously optimized using adversarial loss to obtain the generated vegetation coverage image that meets preset conditions.

2. The method for generating full coverage data of surveying and mapping geospatial space based on machine learning according to claim 1, characterized in that: The historical forest vegetation images include historical NDVI images; the season labels include spring, summer, autumn and winter; and the climate data include temperature, precipitation, humidity and sunshine duration.

3. The method for generating full coverage data of surveying and mapping geospatial space based on machine learning according to claim 1, characterized in that: The vegetation cover image generation model is constructed through a CGAN model, including: a generator, a discriminator and conditional data; the generator includes: the historical forest vegetation image analysis model and the climate data analysis model; the discriminator includes: the generated vegetation cover image authenticity scoring model; the conditional data includes: the historical forest vegetation image, the climate data and the season label.

4. The method for generating full coverage data of surveying and mapping geospatial space based on machine learning according to claim 1, characterized in that: The historical forest vegetation image analysis model is constructed using a CNN model, comprising: a first input layer, a spatial feature extraction layer, and a first output layer; the analysis process of the historical forest vegetation image is as follows: The first input layer inputs the historical forest vegetation image into the historical forest vegetation image analysis model; The spatial feature extraction layer uses a CNN model to extract spatial features of the historical forest vegetation image; The first output layer outputs the spatial features extracted by the spatial feature extraction layer to obtain the forest vegetation spatial features.

5. The method for generating full coverage data of surveying and mapping geospatial space based on machine learning according to claim 1, characterized in that: The climate data analysis model is constructed using an LSTM model, including a second input layer, a time feature extraction layer, and a second output layer. The analysis process of the climate data is as follows: The second input layer inputs the climate data into the climate data analysis model; The time feature extraction layer extracts the time features of the climate data through a double-layer LSTM; The second output layer outputs the time features extracted by the time feature extraction layer to obtain the climate time features.

6. The method for generating full coverage data of surveying and mapping geospatial space based on machine learning according to claim 1, characterized in that: Performing nonlinear processing on the forest vegetation spatial characteristics, the climate temporal characteristics and the season label vector and fusing them to obtain the conditional characteristics; Processing the conditional features to obtain the generated vegetation cover image; the specific process includes: a feature fusion layer and a vegetation cover image generation layer; The feature fusion layer fuses the forest vegetation spatial features, the climate temporal features, and the season label vector through a fully connected layer to obtain the conditional features. The formula for the conditional features is: F fusion =σ1(ω1·g1(NDVI)+ω2·g2(f climate )+ω3·g3(f season )+ω4·g4(NDVI,f climate ,f season ))+b1; Among them, F fusion is the conditional characteristic; NDVI is the spatial characteristic of forest vegetation; f climate is the climate time characteristic; f season is the seasonal label vector; ω1 is the weight of the nonlinear processing of forest vegetation spatial characteristics; g1 is the nonlinear transformation function of forest vegetation spatial characteristics; ω2 is the weight of the nonlinear processing of climate event characteristics; g2 is the nonlinear transformation function of climate time characteristics; ω3 is the weight of the nonlinear processing of seasonal label vector; g3 is the nonlinear transformation function of seasonal label vector; ω4 is the weight of the nonlinear processing of the interaction term of forest vegetation spatial characteristics, climate time characteristics and seasonal label vector; g4 is the nonlinear transformation function of the interaction term of forest vegetation spatial characteristics, climate time characteristics and seasonal label vector; σ1 is the activation function; b1 is the bias term; Combining the conditional features with noise to obtain vegetation cover image features; The vegetation coverage image generation layer uses deconvolution to generate a high-resolution image according to the vegetation coverage image features to obtain the generated vegetation coverage image.

7. The method for generating full coverage data of surveying and mapping geospatial space based on machine learning according to claim 1, characterized in that: The authenticity scoring model for generating vegetation cover images includes: calculating the forest vegetation spatial feature difference, climate time feature difference, and seasonal label difference between the real vegetation cover image and the generated vegetation cover image, as well as the interaction term difference between the three, to obtain the authenticity score of the generated vegetation cover image; the formula for the authenticity score of the generated vegetation cover image is: S=σ2·S1+b2; Among them, S1 is the difference between the real vegetation cover image and the generated vegetation cover image; NDVI real The spatial characteristics of forest vegetation are real vegetation cover images; NDVI gen To generate forest vegetation spatial characteristics for vegetation cover images; f climate,real is the climate characteristics of the real vegetation cover image; f climate,gen To generate climate characteristics of vegetation cover images; f season,real is the season label of the real vegetation cover image; f season,gen is the seasonal label for generating vegetation cover images; α1 is the weight of the difference in forest vegetation spatial characteristics; h1 is the difference in forest vegetation spatial characteristics; α2 is the weight of the difference in climate time characteristics; h2 is the difference in climate time characteristics; α3 is the weight of the difference in seasonal labels; h3 is the difference in seasonal labels; α4 is the weight of the interaction term among forest vegetation spatial characteristics, climate time characteristics and seasonal label vectors; h4 is the difference in the interaction term among forest vegetation spatial characteristics, climate time characteristics and seasonal label vectors; S is the authenticity score; σ2 is the activation function; and b2 is the bias term.

8. The method for generating full coverage data of surveying and mapping geospatial space based on machine learning according to claim 1, characterized in that: The preset condition is that the authenticity score of the generated vegetation coverage image is higher than the authenticity score threshold, and the generated vegetation coverage image meets the condition data.

9. A system for generating full coverage data of surveying and mapping geospatial space based on machine learning, characterized by: include: The data acquisition module is used to collect historical remote sensing image data of the forest at different time points to obtain historical forest vegetation data; The historical forest vegetation data includes: historical forest vegetation images and season labels; collecting time series data related to climate to obtain climate data; collecting actual forest vegetation cover images to obtain real vegetation cover images; The vegetation cover image generation module is used to construct a vegetation cover image generation model based on the historical forest vegetation image, the climate data and the season label to generate a vegetation cover image; the specific steps are: Constructing a historical forest vegetation image analysis model, analyzing the historical forest vegetation image, and obtaining forest vegetation spatial characteristics; Constructing a climate data analysis model, analyzing the climate data, and obtaining climate time characteristics; fusing the forest vegetation spatial features, the climate temporal features, and the season label vector to obtain conditional features; Processing the conditional features to obtain the generated vegetation cover image; A generated vegetation cover image authenticity scoring module is used to construct a generated vegetation cover image authenticity scoring model, and obtain a generated vegetation cover image authenticity score by calculating the difference between the generated vegetation cover image and the real vegetation cover image in combination with the conditional features; The generated vegetation cover image optimization module is used to continuously optimize the authenticity of the generated vegetation cover image using adversarial loss according to the authenticity score of the generated vegetation cover image, so as to obtain the generated vegetation cover image that meets preset conditions.

Images